CMSC828L Spring 2010: Advanced Machine Learning: Combining Statistical and Logical Approaches

Course Description

This course will survey current machine learning topics in the areas of statistical relational learning, structured prediction, and probabilistic logic programming.  It will include some tutorial material on basics of logical and probabilistic learning/reasoning, and then we will read materials from books and papers covering current topics.

The seminar will be very interactive and collaborative. The topics covered and the depth of coverage will depend on the participants' input and interests.

The goal of the course is to give you an overview of current topics in machine learning which applying statistical and probabilistic modeling to complex, richly structured data. The methods are applicable in many areas such as bioinformatics, computer vision, computational linguistics, databases, program analysis, networks and systems. Of particular interest will be the analysis of graph and network data. Along the way, you will pick up some practical experience in reading and presenting research papers, hands-on experience with some of the existing tools, and do a course project that will lead (ideally) to a publishable paper.

In tandem with the course, throughout the semester there will be several invited speakers presenting current work in statistical relational learning and structured prediction. Some of these will be during the scheduled course time, while others, due to schedule constraints, will be outside the regular course time. Students are highly encouraged to attend the invited talks and meet with the speakers.

Prerequisites: Background in machine learning, knowledge representation, and graphical models suggested. Mathematical maturity and a basic course in probability required.

Course Format

This is a seminar course. Each class will consist of presentations and discussion. Students will be required to present papers (25%), explore existing tools (20%) and a class project for the course (40%) . A significant portion of the grade will be based on class participation, which includes class discussion, contributions to the wiki, and demonstrations (15%).

Because of the interactive nature of the course, auditing is discouraged. If you would still like to sit in on the course, see me first, and be prepared to keep up with the reading and participate actively in the class.

Course Credit

This course does not count as a PhD Core or MS Comps course. This course can be used toward PhD coursework as part of the non-core classes required or towards MS coursework (but it is not an MS qualifying course).

Course Material

The required text book is:

Introduction to Statistical Relational Learning, Edited by Lise Getoor and Ben Taskar
Published by The MIT Press. Available at the bookstore here

Supplemental Texts (not required):

Probabilistic Inductive Logic Programming, Edited by De Raedt, L.; Frasconi, P.; Kersting, K.; Muggleton, S.H. Published by Springer.

Predicting Structured Data, Edited by Gökhan H. Bakir , Thomas Hofmann , Bernhard Schölkopf , Alexander J. Smola , Ben Taskar and S. V. N. Vishwanathan Published by The MIT Press

Course Information

Time: Fri 3:00-5:30pm in room: CSI 2120
Professor: Lise Getoor - getoor AT cs.umd.edu
Office hours: TBA in AVW 3217
Web site: http://www.cs.umd.edu/class/spring2010/cmsc828l/

Course Wiki

The class wiki, http://linqs.cs.umd.edu/cmsc828l/, is for students enrolled in the course to share material and discuss content. (coming soon)

Course Mailing List

Tthe class mailing list is for announcements relevant to the class. If you are enrolled in the class please sign up here https://mailman.cs.umd.edu/mailman/listinfo/cmsc828l

Schedule / Syllabus (Subject to Change)

Date
Topic / Papers
Notes
1/29

Introduction

recommended: ch.1, ch.2 and ch.3

slides

2/5
snow day

 

2/12
snow day

please watch:
http://videolectures.net/icml08_domingos_ipk/

2/19

Conditional Random Fields and Factorie

guest speaker: Karl Schultz, UMass

ch. 4, An Introduction to Conditional Random Fields for Relational Learning, Charles Sutton & Andrew McCallum

FACTORIE: Probabilistic Programming via Imperatively Defined Factor Graphs
A. McCallum , K. Schultz , S. Singh, NIPS 2009

slides

2/26

Markov Logic Networks

guest speaker: Lily Mihalkova

 

ch. 12

slides

3/5
Probabilistic Relational Models and Relational Markov Networks

ch. 5

slides

due: MLN exercise

3/12
PRMs and RMNs cont.

ch. 6

slides

due: Project paragraph

3/19
SPRING BREAK
3/26
PSL & GAIA

papers sent to the mailing list

4/2

Alphabet Soup - session 1

due: Project 1 page proposal
4/16

Alphabet Soup - session 2

4/23

Alphabet Soup - session 3

4/30

Structured Prediction I

5/7

Applications -- SRL for the Web

TBA

POSTER SESSION